Laguna 2.1Laguna 2.1

<p class="muted">该发布者未提供说明。以下为该仓库的客观元数据（规格、许可、文件体积、更新时间），不含本站的主观解读。</p>

<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>

定位：老牌量化者，档位命名规范、说明文档最完整。

• 特点：
• 每档位都写清「这个档适合什么场景」，不是只丢一堆文件
• 命名遵循 llama.cpp 约定，Q4_K_M 就是社区通义的那个 Q4_K_M
• 同一个模型常常同时提供静态量化与 imatrix 量化两个目录

怎么选他的包：看仓库里的 README 分档说明，那是少数把「为什么这档质量损失小、那档损失大」写清楚的地方。

注意：他的仓库名常带 Qwen_ Meta-Llama_ 这类前缀（因为 HF 仓库名不允许重复，量化者要加模型组织名消歧），搜仓时别漏掉前缀。

发布页：https://huggingface.co/bartowski

Position: A long-standing quantiser with the most complete tier naming and documentation.

• Characteristics:
• Every tier states what scenario it suits, rather than just dumping files
• Naming follows the llama.cpp convention — Q4_K_M is the community's Q4_K_M
• The same model often ships both static and imatrix directories

How to choose their builds: read the tier-by-tier README in the repo — it is one of the few places that actually explains *why* one tier loses little quality and another loses more.

Note: repo names often carry a Qwen_ or Meta-Llama_ prefix (Hugging Face repo names must be unique, so quantisers prefix the organisation name to disambiguate) — do not miss the prefix when searching.

Release page: https://huggingface.co/bartowski

定位：llama.cpp 自己的官方量化发布渠道。

为什么重要：ggml 是 llama.cpp 的底层张量库，ggml-org 发的 GGUF 由项目核心作者把关。遇到新架构（比如新出的 MoE 稀疏模型），ggml-org 通常是第一个支持并发布可用 GGUF 的。

• 特点：
• 档位偏保守，优先保证能正确加载与推理不出错
• 覆盖不一定全（作者精力有限，只跟进自己关心的模型）
• 适合作为「该架构能不能在 llama.cpp 上跑」的权威参考

适合谁：想第一时间在新架构上尝鲜、或者需要确认某个模型是否已被 llama.cpp 支持。

发布页：https://huggingface.co/ggml-org

Position: The official quantisation release channel of llama.cpp itself.

Why it matters: ggml is llama.cpp's underlying tensor library, and ggml-org's GGUFs are vetted by the project's core authors. When a new architecture appears (a new sparse MoE, say), ggml-org is usually the first to support it and publish a usable GGUF.

• Characteristics:
• Tiers lean conservative, prioritising correct loading and inference that does not break
• Coverage is not necessarily complete (limited authorial bandwidth — they track only models they care about)
• Serves as the authoritative reference for "can this architecture run on llama.cpp at all"

Who it suits: people who want to try a new architecture first-hand, or need to confirm whether llama.cpp supports a given model.

Release page: https://huggingface.co/ggml-org

<p class="muted">该发布者未提供说明。以下为该仓库的客观元数据（规格、许可、文件体积、更新时间），不含本站的主观解读。</p>

<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>

定位：Mirostat 采样算法作者，长期做低比特实验性量化。

• 特点：
• 愿意发别人不敢发的极低比特档位（IQ2_XXS、Q2_K 这类）
• 关注点在「量化方法本身」——会试不同的 imatrix、不同的重要性矩阵来源
• 文件命名带 i1 h2 之类的后缀，标明用的是哪一版重要性矩阵

适合谁：在做「极小显存里塞下最大模型」这类实验，或者想对比不同重要性矩阵对质量的影响。

不适合谁：日常使用。极低比特档位的质量损失通常已经很明显了，除非你确实没有别的选择，否则不建议作为主力。

发布页：https://huggingface.co/mradermacher

Position: Author of the Mirostat sampling algorithm, long-running low-bit experimental quantisations.

• Characteristics:
• Willing to publish very low-bit tiers others dare not (IQ2_XXS, Q2_K and the like)
• Focused on the quantisation method itself — tries different imatrix files and different sources for the importance matrix
• File names carry suffixes like i1 or h2, marking which version of the importance matrix was used

Who it suits: experiments along the lines of "fit the largest possible model into the smallest possible VRAM", or comparisons of how different importance matrices affect quality.

Who it does not suit: everyday use. Quality loss at those very low bits is usually already noticeable — unless you genuinely have no alternative, do not make it your primary build.

Release page: https://huggingface.co/mradermacher

定位：LM Studio 桌面客户端的官方社区组织，为其 GUI 供包。

• 特点：
• 档位偏保守务实，以「在 LM Studio 里能顺畅跑起来」为目标筛选
• 文件命名与 LM Studio 的模型库索引对得上，下载后能被客户端直接识别
• 也会带 MTP / 投机采样相关的特殊版本

适合谁：如果你用 LM Studio 而不是自己敲 llama-server 命令行，这个组织的包省心程度最高。

不适合谁：如果你要精细控制加载参数（tensor split、n_gpu_layers、上下文分配），社区包往往不针对你的硬件调优，仍要自己调。

发布页：https://huggingface.co/lmstudio-community

Position: The official community organisation behind the LM Studio desktop client, supplying packages for its GUI.

• Characteristics:
• Tiers lean conservative and pragmatic, selected for "runs smoothly inside LM Studio"
• File naming lines up with LM Studio's model library index, so downloads are recognised by the client directly
• Also ships special builds related to MTP / speculative sampling

Who it suits: if you use LM Studio rather than driving llama-server yourself, packages from this organisation save the most fuss.

Who it does not suit: if you need fine control over load parameters (tensor split, n_gpu_layers, context allocation) — community packages are usually not tuned to your hardware, so you still tune them yourself.

Release page: https://huggingface.co/lmstudio-community